CN109155018B

CN109155018B - MPC with unconstrained dependent variables for KPI performance analysis

Info

Publication number: CN109155018B
Application number: CN201780028590.0A
Authority: CN
Inventors: A.J.特伦查德; A.奥登-斯威夫特
Original assignee: Honeywell International Inc
Current assignee: Honeywell International Inc
Priority date: 2016-07-28
Filing date: 2017-07-12
Publication date: 2024-02-13
Anticipated expiration: 2037-07-12
Also published as: WO2018022298A1; JP2019523502A; EP3491592A1; EP3491592A4; JP7138621B2; CN109155018A; US10643167B2; US20180032940A1

Abstract

A method (100) of Key Performance Indicator (KPI) performance analysis. A dynamic Model Predictive Control (MPC) process model for an industrial process is provided (101), the industrial process including a Controlled Variable (CV) and a Measured Variable (MV) for an MPC controller. The MPC process model includes at least one KPI that is also included in a business KPI monitoring system for an industrial process. Future trajectories of the KPIs and Steady State (SS) values of the KPIs are estimated (102). Future trajectory and SS values are used (103) to determine the dynamic relationship between key plant operating variables selected from the CVs and MVs and KPIs. The performance of the KPI is analyzed (104), including identifying at least one cause of excess performance or performance problems during operation of the industrial process from the dynamic relationship and the current value of at least a portion of the MV.

Description

MPC with unconstrained dependent variables for KPI performance analysis

Technical Field

The disclosed embodiments relate to Model Predictive Control (MPC) including key performance indicators.

Background

Process control systems are commonly used to manage process facilities that operate physical processes that process materials, such as manufacturing plants, chemical plants, and refineries (oil refineries). Valves, pumps, motors, heating/cooling equipment and other industrial equipment typically perform the actions required to process materials in a processing facility. Process control systems often manage, among other functions, the use of industrial equipment in a processing facility.

In conventional process control systems, controllers are typically used to control the operation of industrial equipment in a processing facility. The controller may monitor the operation of the industrial equipment, provide control signals to the industrial equipment, and/or generate an alarm when a fault is detected. Process control systems typically include one or more process controllers and input/output (I/O) devices communicatively coupled to at least one workstation and to one or more field devices such as via analog and/or digital buses. The field devices may include sensors (e.g., temperature, pressure, and flow rate) sensors), as well as other passive and/or active devices. The process controller may receive process information, such as field measurements made by field devices, to implement a control routine. Control signals may then be generated and sent to the industrial device to control the operation of the process.

Industrial plants typically have control rooms with displays for displaying process parameters such as key temperature, pressure, fluid flow and flow levels (flow levels), key valves, operating positions of pumps and other equipment, etc. Operators in the control room may control various aspects of plant operation, typically including override (override) automatic control. Typically in a plant operating scenario, an operator desires operating conditions such that the plant is always operating at its "best" operating point (i.e., the benefit associated with the process is at a maximum value, which may correspond to the amount of product generated) and thus approaching an alarm limit. Based on changes in the composition of the raw materials used in the chemical process, changes in product requirements or economics, or other changes in constraints, the operating conditions can be changed to increase the benefit. However, due to variability in the process, there is an increased risk associated with operating the plant closer to the alarm limit.

Advanced process controllers implement multivariable Model Predictive Control (MPC), an Advanced Process Control (APC) technique used to control the operation of devices running industrial processes. The model is a set of independent variables and a generally linear dynamic relationship between the independent variables. The model may take different forms, with Laplace transforms and ARX models being conventional model implementations. Nonlinear relationships between variables are also possible.

MPC control techniques typically involve analyzing received current input (e.g., sensor) data using empirically derived process models (i.e., based on historical process data), where the models identify how industrial equipment should be controlled (e.g., by changing actuator settings) and thus operate based on the received input data. The control principle of MPC uses three (3) class process variables, the manipulated variable MV and some measured disturbance variables (disturbance variable) (DV) as independent variables, and the Control Variable (CV) as dependent variables. The model includes a response of each CV to MV/DV changes, and predicts future effects on the CV based on changes in MV and DV.

In many industrial and commercial customer applications, key Performance Indicators (KPIs) are used by business KPI monitoring systems to track whether an enterprise or organization is performing for acceptable criteria, such as in compliance with law, productivity, energy usage, and maintaining product or quality of service and profitability. There are typically a wide variety of KPIs, for example, from lost work time from operators due to accidental injury, repair shop performance, environmental emissions up to productivity, quality variation, and energy and chemical consumption.

Some KPIs used by business KPI monitoring systems are independent of variables within the scope of the MPC controller (e.g., lost time injuries and repair shop performance KPIs). However, KPIs associated with production targets such as feed rate (feed rate) of a process, productivity of various products, product 1 versus yield (product yield) 2, energy consumption, etc., will typically overlap significantly with the CV, MV, or DV of the MPC model. In some cases, the same variables used to calculate the KPIs are also configured as MPC models CV or MV (because MPC controls and optimizes important production variables). In other cases, KPIs will be highly correlated with MPC MVs and DVs, and thus KPIs may be predicted/planned using the same MPC tools and workflow. This may include specific energy usage or yield that may be used for performance monitoring.

Disclosure of Invention

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description that includes the drawings. This summary is not intended to limit the scope of the claimed subject matter.

The disclosed embodiments recognize that while KPIs and time-based trends of KPIs are useful for tracking performance and helping to quickly identify changes in process performance in an industrial process, further detailed analysis is often required to understand the root cause of KPI performance changes in order to correct, mitigate, or utilize changes in one or more KPIs. Data science and data analysis are growing areas and there are commercially available universal toolsets to support practitioners in this area. However, analyzing the causes of poor KPI performance can be challenging because the visibility or inference of root cause events that can affect a given KPI, as well as the potential diversity of those root cause events, may be measured in disparate systems or not measured at all. Another generally important factor is the effect of closed loop control in the measured system, especially when the control can be switched between active (closed loop) and inactive (open loop) modes, being multiple, i.e. having multiple inputs (CVs, DVs) and multiple outputs (MVs), which can be incorporated into or taken out of the closed loop control, or non-square (i.e. the number of CVs does not match the number of MVs).

It is also recognized that for Model Predictive Control (MPC), there are problems caused by correlations between various process variables in the MPC process model for an industrial process that change over time, depending on whether the control system is active or partially active, and whether a particular set of CVs and MVs are within the active set of controls. The disclosed embodiments also recognize that it is much simpler to implement KPI analysis as part of a closed loop process control application (in comparison to traditional "general" data analysis methods, where statistical regression and clustering (cluster) techniques are used to analyze large sets of historical time series data, but do not contain models of the process and behavior of the control layer). While it is possible to perform KPI analysis on an MPC application for purposes of evaluating whether an MPC controller is functioning well, being properly configured, and being effectively used by an operator, the user herein is an MPC maintenance engineer/MPC team leader. Furthermore, rebinding (typing) of MPC KPI analysis back to the traffic KPI monitoring system is not known. There is no idea of including KPI Unconstrained Dependent Variables (UDVs) in the MPC model without upper or lower control limits.

In many industrial sectors (such as, for example, the process and chemical industries), closed loop control, in particular MPC control, is commonly used. Closed loop control is typically configured to control and enhance the benefits of multiple CVs by adjusting MVs that are used directly as KPIs or that typically strongly affect other KPIs. Applications have some type of process model that describes the behavior of the system. Modern MPC control applications are predictive to provide early real-time indications of CVs and future trajectories (trajectories) of MVs, and with available MVs and their configured high and low limits, whether MPC will be able to control CVs with specified CV high and low limits, or whether these limits will be violated. The control application relatively quickly performs monitoring real-time constraints of the monitored control system and reconciles the CV predictions with the current process measurements (recorcile). If KPIs have been included as CVs or MVs in the MPC model and/or added as KPIs UDVs, MPCs may be used to predict future trajectories of KPIs and whether they will deviate from their targets configured in the business KPI system.

The disclosed embodiments include a method of KPI performance analysis that includes providing a dynamic MPC process model for an industrial process that includes a plurality of MVs and a plurality of CVs for an MPC controller implemented by a processor having a memory storing the MPC process model. The MPC process model includes at least one KPI that is also included in a business KPI monitoring system for an industrial process. The future trajectory of the KPI and the Steady State (SS) value where the KPI will stabilize are estimated. Future trajectory and SS values are used to determine the dynamic relationship between key plant operating variables selected from the plurality of CVs and the plurality of MVs and KPIs. Analyzing the performance of the KPI, including identifying at least one cause of excess performance or performance problems during operation of the industrial process from the dynamic relationship and the current value of at least a portion of the MV.

Drawings

FIG. 1 is a flowchart illustrating steps in an example method of a method of KPI performance analysis according to an example embodiment.

FIG. 2 is a block diagram of an example process control system including an MPI controller implementing the disclosed MPC control

FIG. 3 is an example simulated flow diagram for a debutanizer process according to an example embodiment.

FIG. 4 illustrates an example MPC control schematic for a debutanizer process showing an example MPC control strategy employed with multiple KPI UDVs, according to an example embodiment.

Detailed Description

The disclosed embodiments are described with reference to the drawings, wherein like reference numerals are used to refer to like or equivalent elements throughout. The figures are not drawn to scale and they are merely provided to illustrate certain disclosed aspects. Several disclosed aspects are described below with reference to example applications for illustration. It should be understood that numerous specific details, relationships, and methods are set forth to provide a full understanding of the disclosed embodiments.

One of ordinary skill in the relevant art, however, will readily recognize that the subject matter disclosed herein may be practiced without one or more of the specific details or with other methods. In other instances, well-known structures or operations are not shown in detail to avoid obscuring certain aspects. The present disclosure is not limited by the illustrated ordering of acts or events, as some acts may occur in different orders and/or concurrently with other acts or events. Moreover, not all illustrated acts or events are required to implement a methodology in accordance with the embodiments disclosed herein.

The disclosed embodiments implement certain types of operational KPIs with an MPC control and optimization framework enabled by alignment between KPI management activities and MPC targets, recognizing that it is difficult to analyze the cause of poor KPI performance unless MPC performance and configuration are considered. FIG. 1 is a flowchart illustrating steps in an example method 100 of KPI performance analysis according to an example embodiment. Step 101 includes providing a dynamic MPC process model including a plurality of CVs and a plurality of MVs for an industrial process of an MPC controller implemented by a processor having a memory storing the MPC process model.

The MPC process model includes at least one KPI that is also included in a business KPI monitoring system for an industrial process. The KPIs may include KPI UDVs, which have no upper or lower control limitations, as described above. The Disturbance Variable (DV) is an independent variable despite the lack of a relative control limit. The MPC model may comprise a plurality of KPI UDVs. Known KPIs include feed rate (feed rate) and product flow, and disclosed KPIs UDVs may include key production and unit performance monitoring variables such as production, energy and chemical consumption, and plant efficiency. KPIs are typically calculated externally. The calculations may be implemented in a variety of ways, such as within an MPC encoding calculation or within a DCS system. The calculation may be a simple ratio of streams (e.g., yield), thermal equilibrium, or more complex correlation.

KPIs are typically, but not limited to, calculated values. Feed and product streams are examples of simple directly measured KPIs. For example, the yield is a calculated value, the product stream divided by the feed stream. Energy efficiency, specific energy consumption (specific energy consumption), plant efficiency is a value calculated from other directly measured variables. KPI values will typically be calculated for the traffic KPI monitoring system. However, there is also a need to calculate KPI values for MPC applications (which values may then be made available to the business KPI monitoring system to avoid re-work). For MPCs, there are typically two options to implement the calculations, either the MPC itself or a calculation block in a distributed control system.

As known in the art, MPC models are defined in terms of open loop process behavior, but MPC is used for closed loop control by a "reverse" model. The net effect is the process behavior change when the MPC turns on to reflect the closed loop behavior.

Step 102 includes estimating future trajectories of KPIs and SS values where KPIs will stabilize. Step 103 includes using the future trajectory and SS values to determine a dynamic relationship between key plant operating variables selected from the plurality of CVs and the plurality of MVs and the KPIs.

Step 104 includes analyzing the performance of the KPI, including identifying at least one cause of the performance or problem of performance exceeding during operation of the industrial process from the dynamic relationship and the current value of at least a portion of the MV and optionally one or more of the CVs. Analysis may be performed when the MPC controller is on, off or partially on. A significant advantage of the disclosed embodiments is that KPI analysis takes into account MPC status and how it (MPC status) affects the behavior of an industrial process. Analyzing performance of KPIs may be performed by an MPC controller or a separate computing device implementing the disclosed algorithm, including in one particular embodiment a cloud-based computer.

Identifying the cause of the problem may include identifying which of a plurality of MVs are causing a change in KPI. The method may further comprise providing the results of the analysis to a business KPI monitoring system, and a user of the business KPI monitoring system typically uses the results of the analysis on a display screen displaying a dashboard (dashboard) view to address (troublesboost) performance problems or in some cases reasons for excess performance. The business KPI monitoring system user may suggest a change in the setting of at least one MPC model parameter selected from MV and CV, or initiate a query by the process operator to find out why a given KPI is currently limited or they may trigger a workflow investigation for another person.

The method may further include updating the dynamic relationship based on the analysis of the performance. Future trajectories may be used for automatic alarms and event detection by operators.

Most commercial MPC software includes optimizers to direct the MPC controller to an operating point that maximizes the benefit. The optimum point is found by defining economic values or costs for one or more CVs and MVs. The optimizer calculates ideal operating points within the high and low bounds. The optimizer basically gives the MPC its own ideas and can drive the MPC towards or away from KPI targets or weigh one KPI against another. In the case where the MPC process model includes an optimizer, the method may further include identifying a cause of the problem when the optimizer is biasing the KPI away from its target.

KPIs may be affected by independent variables, such as MV or DV, that are not currently in the MPC range. However, other process variables may be searched using the MPC configuration tool to determine if they have a measurable effect on the KPI UDV. If this is the case, they may be added to the scope of the MPC controller so that the method may further include searching the industrial process for at least one other variable affecting the KPI that is not included in the MPC process model, and adding the other variable to the MPC process model.

Predictability of KPI UDV based on MPC model can be analyzed. A good (i.e. predictable) MPC model means that further analysis can be continued. The poor model means that there are other important factors affecting KPI UDV, which should generally be explored. If the model prediction is bad, a workflow may be triggered to explore what this might be, for example, with tools for historical model identification. The quality of KPI predictions will be based on step 102 (estimating future trajectories) in which future KPI values are predicted. These predictions will then be stored (at future time stamps) and then compared to the measured values (when the actual time has advanced to the corresponding time stamp). An additional "test" model may be added to evaluate whether the prediction improves. This helps to improve understanding of key contributors to KPIs. Examples of factors that lead to poor KPI predictions are that an existing MPC model is outdated and needs to be updated to reflect recent changes in process behavior, and that an MPC model needs to be augmented to include additional plant information. Both of these effects can be overcome by a combination of commercially available factory step test tools (e.g., honeywell (PROFIT STEPPER) and historical model identification tools).

If MPC model predictions are found to be reliable from the analysis step, the predicted KPI values are generally useful as a guideline for the process operator. The MPC model and MPC constraints can then be analyzed to understand the mostly negatively affected (impactive) controlled/uncontrolled variables about KPI variables, as well as whether the KPI is being "suppressed" (constrained) by process over-constraints (e.g., conservative MV or CV limits) and how much additional revenue (extra) can be achieved if certain limits are relaxed, which quantifies the incremental improvement of KPI if the associated MPC CV and MV limits are slightly relaxed.

The challenge is how to provide a good "line of sight" between the time aggregated KPI (e.g., one shift (shift) of operation or on the day) and the actions taken by the operator. For example, it is possible to evaluate which individual MV and CV constraints are suppressing a given KPI for its target value (e.g., by relaxing MPC controller limits to ideal range limits specified by a process/reliability engineer). This may be an important job to be performed on each controller iteration (execution cycle-per-minute), and for KPIs this typically makes doing so more reasonably less frequently, e.g., each controller to steady state time (Controller Time To Steady State) (TTSS). One approach is to average the actual values of MV and CV steady state values over this period of time and use these as the starting point for the "optimization" case in an optimization tool, such as EXCEL or Honeywell PROFIT CONTROLLER optimization solver, relax the average limit and evaluate the effect on KPI variables. The influence of DV can also be considered.

Additional KPI variables not used in conventional MPC models are integrated into MPC applications as Unconstrained Dependent Variables (UDV) that do not have upper or lower control limitations so that they are not controlled by the MPC. As described above, there is typically good alignment between variables included in an MPC control application (or an APC in the process industry) and a number of operation-related KPI variables, such as productivity, yield, product quality and energy usage. Good alignment is provided because many important production variables will typically be monitored by the business KPI system and controlled and manipulated by the MPC controller. There are other variables that are monitored by the business KPI system that are highly correlated to the MPC variables or may be correlated to a combination of MPC variables. This is because MPC is typically demonstrated by improvement in operating performance.

However, some specific KPI variables are not included in conventional MPCs because they may be considered as duplicate information or have no specific control or optimization objectives. If these KPI variables are affected by at least one MV in the MPC model, it is recognized that they may be included as UDVs in the MPC model run by the MPC controller, providing several significant benefits at additional engineering costs, which are typically very small. Benefits include computing KPI plans in the MPC based on MVs, root cause analysis, including identifying which MV changes cause KPI changes, and KPI plans for alarm and early event detection, such as when transient bias is excessive or steady state predictions are considered far from their targets.

Referring now to FIG. 2, a process control system 200 is shown in which a disclosed MPC process controller 211 having KPIs including at least one KPI UDV for KPI performance analysis is communicatively coupled to an MPC server 228, a data historian (History) 212, and to one or more host workstations or computers 213 (which may comprise Personal Computers (PCs), workstations, etc.) each having a display screen 214 via a network 229. The MPC process controller 211 includes a processor 211a and a memory 211b.

The control system 200 also includes a fourth level (L4) comprising an L4 network 235 with workstations or computers 241 and a KPI traffic monitoring system including a traffic monitoring system server 240. The traffic monitoring system server 240 is connected to the network 229 through a firewall 236.

The MPC controller 211 is also connected to the field devices 215-222 via input/output (I/O) device(s) 226. The data historian 212 may generally be any type of data collection unit having memory and software, hardware or firmware for storing data, and may be separate from or part of one of the workstations 213. The MPC controller 211 is communicatively connected to the host computer 213 and the data historian 212 via, for example, an ethernet connection or other communication network 229.

The communication network 229 may be in the form of a Local Area Network (LAN), wide Area Network (WAN), telecommunications network, etc., and may be implemented using hardwired or wireless technology. The MPC controller 211 is communicatively connected to the field devices 215-222 using hardware and software associated with, for example, standard 4-20ma devices and/or any smart communication protocol, such as, for exampleFieldbus protocol (Fieldbus)>Protocol, wireless HART ^TM Protocols, etc.

The field devices 215-222 may be any type of device, such as sensors, valves, transmitters, positioners, etc., while the I/O device 226 may generally conform to any communication or controller protocol. The field devices 215-218 may be standard4-20ma device orDevices that communicate with the I/O devices 226 over analog lines or combined analog/digital lines, while the field devices 219-222 may be "smart" field devices, such as Fieldbus field devices, that communicate with the I/O devices 226 over digital buses using Fieldbus protocol communications.

The MPC controller 211, which may be one of many distributed controllers within the plant 205, has at least one processor therein that implements or oversees one or more process control routines, which may include a control loop stored therein or otherwise associated therewith. The MPC controller 211 also communicates with the devices 215-222, the host computer 213, and the data historian 212 to control the process. The process control element may be any component or section of a process control system including, for example, routines, blocks or modules stored on any computer readable medium and executed by a processor such as the CPU of a computer device.

The control routines, which may be any portion or module of a control process such as a subroutine, a portion of a subroutine (e.g., lines of code), etc., may generally be implemented in any software format, such as using ladder logic (ladder logic), sequential function charts, function block diagrams, object oriented programming or any other software programming language or design paradigm. Likewise, the control routines may be hard-coded into, for example, one or more EPROMs, EEPROMs, application Specific Integrated Circuits (ASICs), or any other hardware or firmware elements. The control routine may be designed using a variety of design tools, including graphical design tools or other types of software, hardware or firmware programming or design tools. Thus, the MPC controller 211 may generally be configured to implement control strategies or control routines in a desired manner. In one embodiment, the MPC controller 211 implements a control strategy using what are commonly referred to as function blocks, wherein each function block is part of an overall control routine or object and operates in conjunction with other function blocks (typically via communications called links) to implement process control loops within the process control system 200.

The function blocks typically perform input functions, one of which is associated with, for example, transmitters, sensors or other process parameter measurement devices, and one of which is associated with, for example, executing control routines of the MPC that control the operation of some device (e.g., a valve) within the process control system 200. Function blocks may be stored in the MPC controller 211 and executed by the MPC controller 211, typically these function blocks are used or associated with standard 4-20ma equipment and such asSome type of intelligent field device, such as a device, or may be stored in and implemented by the field device itself, which may be +.>In the case of Fieldbus devices. Still further, the functional blocks implementing the controller routine may be implemented in whole or in part in the host workstation or computer 213 or in any other computer device.

Example

The disclosed embodiments are further illustrated by the following specific examples, which should not be construed as limiting the scope or content of the present disclosure in any way.

For example, consider the simulated flow diagram 300 of the debutanizer process depicted in fig. 3. The debutanizer process is a distillation process common in many refineries that separates light Liquefied Petroleum Gas (LPG) components from a mixed naphtha stream. FIG. 4 illustrates an example MPC control schematic of a debutanizer process showing an example MPC control strategy with adoption of KPIs comprising several KPI UDVs.

As can be seen, many of the key distillation KPIs for the process are naturally included in the control strategy of the MPC model, such as:

unit feed rate as manipulated variable MV1.

Top (overhead) product quality as control variable CV1.

Bottom product quality as controlled variable CV2.

However, a number of KPI parameters have been added to the MPC controller that were not used as KPIs in conventional MPC control models, shown as KPI UDVs 1-4 (being a specific energy usage, reboiler heat of operation (duty-heat) stream, LPG yield and condensate yield (condensing yield)), each of which lacks both high and low control limits. It can be seen in the MPC model that each of the KPIs UDVs has only Steady State (SS) and future values. It is recognized that implementing these KPIs UDV 1-4 as additional KPIs in the MPC model without high or low restrictions would provide at least two benefits:

1. the MPC controller information (such as the values of the other controllers MV, and the CV and operator entered limits) can be used, and the optimizer configuration (controller state) to help diagnose (find root cause) why any corresponding KPI is executing above or below their targets (limits or ranges);

2. future values (trajectories) of KPIs are predicted so that the operator can be automatically alerted in real time if the predicted KPIs (future values) change significantly in a short period of time to enable the preemptive to take possible corrective actions.

Analyzing the cause(s) of poor (or exceeding) KPI performance may be implemented as a multi-step process as described below.

Step 1 prediction quality：

A first step may include evaluating whether the MPC model predicts each KPI value well. Techniques for analyzing the predicted quality of general MPC controlled variables are well established and have been simplified to be implemented in commercially available products such as the PROFIT EXPERT toolset of Honeywell. The method involves the evaluation of predicted MPC model bias relative to movement of variables within range of the MPC controller and other external variables. If the MPC model is bad (KPI deviation is large), the performance of the MPC model may be improved using a model update workflow using step test tools such as PROFIT STEPPER and PROFIT SUITE ENGINEERING STUDIO from Honeywell. The MPC model may be undesirable because it does not reflect changes in the behavior of the process or is incomplete, i.e., does not include all influencing factors. Known tools may be used to search for the influencing process variable(s) from the historical data stored in the data historian 212, which may then be refined using step test tools.

Step 2 determines the root cause(s) of KPI target deviation:

there are a number of consecutive steps that can be followed to determine the cause of a KPI deviating from its target value or range:

step 2a: a check may be made to determine the percentage of time that a limit clamp (KPI) has been entered for an operator or engineer and whether the MPC controller limit is consistent with the overall KPI limit. For example, the feed rate (MV 1) and the two quality KPIs (CV 1 and CV 2) have associated limitations. In this real-time view (a snapshot of the MPC controller at one instant), they are not limited by those limitations. However, over the course of KPI polymerization periods, they may reach their limit for some percentage of time.

If the KPIs within the MPC model are clamped at MPC limits and the MPC limits are inconsistent (i.e., more restrictive) with their KPI targets in the business KPI system, then the mismatch may be automatically marked as a cause of poor KPI performance. For example, poor KPI performance of the debutanizer feed rate may be reported as "the debutanizer feed rate is below the target during the KPI polymerization period. 67% of this deviation can be attributed to the process operator clamping the MPC feed rate to an average of 20m ³ Facts of/hr.

Step 2b: when the KPI is not limited to include KPIUDV (without limitation) limitations within the MPC controller, the MPC controller performance should be checked to determine what is inhibiting the KPI from moving in a direction that meets the aggregated KPI goal. This may be due to:

mpc controller economics (linear program and quadratic program weights) have been configured to move KPIs in the wrong direction (away from aggregated KPI targets);

the MPC controller economics is such that its optimizer has calculated that it can earn more money by moving one or more KPIs in the wrong direction in order to move CVs and MVs (with greater economic value) towards its targets. A simple example is when a treatment plant is sold out, the stimulation to produce more product (consume more feed rate) is stronger than the stimulation to reduce the total energy consumed or even the specific energy consumed. Note, however, that if the primary optimization process (with respect to maximization of productivity) encounters its constraints such that productivity cannot be further increased, a secondary objective such as reducing any incremental specific energy consumption (i.e., reducing any incremental specific energy consumption) may play a role.

3. Another related variable within the controller is constraining the MPC controller from moving KPI(s) in a favorable direction.

By comparing steady state predictions of KPIs (e.g., CVs or MVs in MPCs) with current measurements and by examining objective function economics, the intent of an MPC optimizer can be established in one of two ways.

The optimizer may determine the manner in which to move the free MV by taking into account the effect of direct MV economic weights and the change in MV on the CV with economic weights. This can be calculated analytically for a given optimizer formula, but in a general sense this can be represented by the following equation:

wherein:representing the partial derivative;

Obj _cost is the value of the cost-based objective function:

Obj _cost ＝fn(CV _i,n ,MV _j,m ,Economics)

the predicted CV value is based on the past MVs and DVs as shown in the following equation:

CV _i ＝fn(∑ _j ＝1 _,m g _i,j ×MV _j +∑ _k＝1,l g _i,k ×DV _k )

if the stimulus to the MVi is affirmative, the optimizer will always seek to minimize the MVi value, subject to the MPC controller limitations. If this is a KPI with an aggregate goal to be maximized, the process relies on the process operator to always keep the low limit at or above the aggregate KPI limit in order to achieve the aggregate KPI limit. Otherwise, there will always be a mismatch between MV and KPI.

Likewise, if the KPI UDV is not to be optimized in the correct direction by any application MV, the aggregated KPI objective can typically be achieved by only the correct setting of operator restrictions, which can be verified from the operational data. This covers case 1 above.

The above case 2 can be determined by analyzing whether there is a mix of MVs, some of which will move a given KPI in the correct direction (towards the aggregated KPI direction) and others will move it in the wrong direction. In this case, the optimizer may decide whether to move the KPI UDV towards or away from the aggregated KPI based on the degrees of freedom of the respective MVs to be moved. This can be identified by analyzing the direction of optimization, where MV and CV are constraint and economic weights.

In terms of the active set of constraints, case 3 above can be evaluated by examining the solution returned from the optimization step. Conceptually, this amounts to limiting the predicted MPC variables to where they are (at steady state) in order to determine if the optimizer then moves the KPI towards its aggregated KPI goal.

Step 3: aggregation and thresholding (threshold)

Multiple steps in the analysis use real-time information to evaluate why KPIs are constrained away from their aggregate targets. Typically, these reasons need to be aggregated over a KPI reporting period, arranged according to their percent applicability and the most prominent reasons reported as reasons for KPI performance problems.

While various disclosed embodiments have been described above, it should be understood that they have been presented by way of example only, and not limitation. Many modifications may be made to the subject matter disclosed herein in accordance with the present disclosure without departing from the spirit or scope of the disclosure. In addition, while a particular feature may have been disclosed with respect to only one of several implementations, such feature may be combined with one or more other features of the other implementations as may be desired and advantageous for any given or particular application.

As will be appreciated by one of skill in the art, the subject matter disclosed herein may be implemented as a system, method, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a "circuit," module, "or" system. Furthermore, the present disclosure may take the form of a computer program product embodied in any tangible medium of expression having computer-usable program code embodied in the medium.

Claims

1. A method (100) of Key Performance Indicator (KPI) performance analysis, comprising:

providing (101) a dynamic Model Predictive Control (MPC) process model for an industrial process, the industrial process comprising a plurality of Controlled Variables (CVs) and a plurality of Measured Variables (MVs) for an MPC controller, the MPC controller being implemented by a processor (221 a) having a memory (211 b) storing the MPC process model, the MPC process model comprising at least one KPI, the at least one KPI also being included in a traffic KPI monitoring system for the industrial process, wherein the KPI comprises: at least one first KPI that is not constrained to a limit within the MPC controller and at least one second KPI that is constrained to the limit within the MPC controller;

estimating (102) a future trajectory of the at least one first KPI and a Steady State (SS) value in which the at least one first KPI will stabilize;

determining a dynamic relationship between the key plant operating variables selected from the plurality of CVs and the plurality of MVs and the at least one first KPI using (103) the future trajectories and the SS values, an

-analyzing (104) the performance of the at least one first KPI to identify at least one cause of problems exceeding the performance or the performance during operation of the industrial process from the dynamic relationship and a current value of at least a part of the MV.

2. The method of claim 1, wherein the at least one first KPI comprises a KPI Unconstrained Dependent Variable (UDV) without upper or lower control limits.

3. The method of claim 1, wherein the identifying the cause of the problem comprises identifying which of the plurality of MVs are causing a change to the at least one first KPI.

4. The method of claim 1, further comprising providing results of the analysis to the traffic KPI monitoring system, and a user of the traffic KPI monitoring system utilizing the results of the analysis to solve the problem of the exceeding the performance or the performance.

5. The method of claim 1, further comprising updating the dynamic relationship based on the analyzing the performance.

6. The method of claim 1, further comprising using the future trajectory for automatic alert and event detection.

7. The method of claim 1, wherein the MPC process model includes an optimizer, wherein the identifying the cause of the problem includes identifying when the optimizer deviates the KPI from its target.

8. The method of claim 1, further comprising searching for at least one other variable in the industrial process that affects the KPI that is not included in the MPC process model, and then adding the other variable to the MPC process model.

9. The method of claim 1, further comprising:

if the MPC limit is inconsistent with the KPI target, analyzing the performance of the at least one second KPI clamped at the limit within the MPC controller to flag the mismatch.

10. A Model Predictive Control (MPC) controller, comprising:

a processor having a memory storing at least one algorithm executed by the processor for implementing a dynamic MPC process model of an industrial process running in an industrial plant (205), the industrial process comprising a plurality of Measured Variables (MVs) and a plurality of Controlled Variables (CVs), the MPC process model comprising at least one Key Performance Indicator (KPI) also included in a business KPI monitoring system for the industrial process, wherein the KPI comprises: at least one first KPI that is not constrained to a limit within the MPC controller and at least one second KPI that is constrained to the limit within the MPC controller;

the MPC process model is configured to:

estimating a future trajectory of the at least one first KPI and a Steady State (SS) value in which the at least one first KPI will stabilize;

determining a dynamic relationship between the key plant operating variables selected from the plurality of CVs and the plurality of MVs and the at least one first KPI using the future trajectory and the SS values; and

analyzing the performance of the at least one first KPI includes identifying from the dynamic relationship and a current value of at least a portion of the MV at least one cause of a problem exceeding the performance or the performance during operation of the industrial process.

11. The MPC controller of claim 10, wherein the MPC process model comprises an optimizer, wherein the identifying the cause of the problem comprises identifying when the optimizer deviates the KPI from its target.

12. The MPC controller of claim 10, further comprising analyzing performance of the at least one second KPI clamped at the limit within the MPC controller to flag a mismatch if the MPC limit is inconsistent with the KPI target.